xiuyul / mamba-2.8b-zephyr

mamba-2.8b-zephyr

This model is a fine-tuned version of xiuyul/mamba-2.8b-ultrachat on the HuggingFaceH4/ultrafeedback_binarized dataset trained using Direct Preference Optimization (DPO) .

The base model, xiuyul/mamba-2.8b-ultrachat , was instruction-tuned from state-spaces/mamba-2.8b-slimpj on the HuggingFaceH4/ultrachat_200k dataset.

It achieves the following results on the evaluation set:

• Loss: 0.4996
• Rewards/chosen: -0.4523
• Rewards/rejected: -1.6105
• Rewards/accuracies: 0.7857
• Rewards/margins: 1.1582
• Logps/rejected: -290.1885
• Logps/chosen: -359.0926
• Logits/rejected: 23.0423
• Logits/chosen: 23.1861

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-07
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• total_eval_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 3

Training results

Framework versions

• Transformers 4.35.0
• Pytorch 2.1.1+cu121
• Datasets 2.14.6
• Tokenizers 0.14.1